LEMLEMWORKING PAPER SERIES

Where Do We Go From Here? Market Access and Regional Development in

Italy (1871-1911)

Anna Missiaia °

° Lund University, Sweden

2015/18 June 2015ISSN(ONLINE) 2284-0400

Where Do We Go From Here? Market Access and Regional

Development in Italy (1871–1911)

Anna Missiaia∗

Lund University

Abstract

Italy has been characterized, throughout its history as a unified country, by large

regional differentials in the levels of income, industrialization and socio-economic

development. This paper aims at testing the New Economic Geography hypothesis on

the role of market access in explaining these regional differentials. We first quantify

market access of the Italian regions for benchmark years from 1871 to 1911 following

Harris (1954). We then use these estimates to study the causal link between GDP

per capita and market potential following Head and Mayer (2011). The main result of

this paper is that only domestic market potential, which represents the home market,

shows a “traditional” North-South divide. When international markets are introduced,

the South does not appear to lag behind. Regression analysis confirms that market

potential is a strong determinant of GDP per capita only in its domestic formulation.

This suggests that the home market in this period mattered far more for growth than

the international markets, casting new light on one of the classical explanations to the

North-South divide.

JEL Classification Numbers: N33, N73, N93, O18, R11.

Keywords: Economic Geography; Economic History of Italy; Market Potential.

∗Lund University, Economic History Department, Room 2110, Alfa 1, Scheelevagen 15 B, Lund, Skane

22363, Sweden, e-mail:anna.missiaia@ekh.lu.se. The author is grateful to Brian A’Hearn, Carlo Ciccarelli,

Emanuele Felice, Giovanni Federico, Stefano Fenoaltea, Steve Gibbons and Max Schulze for advice and

suggestions, and to participants of the seminars at the London School of Economics, Bocconi University,

EHS Annual Conference, Carlos III Summer Schools and European Business History Association Summer

School, FRESH Winter School for helpful comments.

1

1 Introduction

Italy’s political Unification dates back to 1861. It was the result of two independence

wars against the Habsburg Empire and the Expedition of the Thousand through which

Southern regions were annexed to the newly unified North. At the time of Unification,

the regions of Italy were not at all homogeneous in terms of their economic structure and

performance. Although they did not show as large a North-South divide in terms of GDP

per capita as today’s, it is in this period that the gap starts widening (see Figure 1).

It is in the 1880s that the Italian regions start to polarize in terms of GDP per capita

and even more sharply in the level of industrial production. During the first decades after

Unification, most of the economic gap was between the Northwestern regions (Piedmont,

Lombardy and Liguria) versus the rest of the country. If we look at the indicators of

economic development, such as the literacy rate, the level of regional inequality appears

evident at the time of Unification. Figure 2 shows the literacy rates for each region in the

benchmark years 1871, 1881, 1901 and 1911. Finally, looking at the industrial value added

per capita (which corresponds specifically to the part of GDP produced by industry) in

Figure 3, the formation of the gap during this period is much more evident, due to the

process of industrialization which was largely concentrated in the Northwestern regions of

Piedmont, Lombardy and Liguria.

The historiography has brought forward several explanations for what is often referred

to as the “Questione Meridionale”, the Southern Question. The hypotheses proposed by

scholars range from colonial exploitation of the South by the North (Villari (1979) and

Sonnino (1877)), to differences in agricultural structure (Cafagna (1989) and Zamagni

(1990)), to differences in the institutional framework (Felice and Vasta (2012)) all the way

to cultural differences (Galassi (2000) and A’Hearn (2000)).1

Another line of research has been focusing on the physical geography of Italy as the

primary cause of the backwardness of the South. For instance, Fenoaltea (2006, p. 261)

gives a possible explanation for the regional patterns of industrialization in Italy before

the First World War based on the comparative advantages of the North, in particular

in terms of energy endowment from water, but he does not translate it into a formal

model. Daniele and Malanima (2007, p. 181) discuss the role of the physical distance

of the southern regions to the centre of Europe and they claim that the position of the

South constituted a natural disadvantage for its industrialization. This explanation is

strongly rejected by Felice (2013, p. 201) who uses several other historical examples, i.e.

1For a more detailed overview of these views see Missiaia (2014).

2

Figure 1: GDP per capita of the Italian regions, 1871–1911 (constant 1911 prices,

Italy=100).

Source: Felice (2009) for 1881 and 1901 and Brunetti et al. (2011) for 1871, 1891,

1901 and 2009.

California, Japan and Southeast Asia, to demonstrate that mere physical distance from

the core does not necessary imply a disadvantage. One other existing piece of research

that links Economic Geography to the regional disparities in Italy is a working paper

by A’Hearn and Venables (2011). The paper explores the relationship between economic

disparities, internal geography and external trade for the 150 years of the unitary history of

Italy. The authors propose different explanations for what drove economic activity across

Italian regions in different periods: in the period 1861–1890 the main driver was natural

advantage; in the period 1890–1950 it was access to the domestic market; and finally in

the post-war period it was the access to foreign markets. The purpose of this paper is

to contribute to the existing debate on the role of geography in the Italian North–South

divide.

In Economic Geography, we know that there are two competing views on why certain

regions attract economic activity more than others. The traditional Heckscher-Ohlin (H-

O) view focuses on factor endowment to explain the location of economic activity, meaning

that regions with a higher endowment of natural, financial or human resources attract

economic activity. Opposite to this, the New Economic Geography (NEG) view considers

market access as the main force. While measuring endowments is a fairly straightforward

procedure, if not in terms of data collection at least in terms of methodology, measuring

3

Figure 2: Literacy rates in the Italian regions, 1871–1911.

Source: A’Hearn et al. (2011).

market access is far more complicated. Reliable measures of market access are an essential

starting point in order to evaluate NEG forces. This paper looks at the role of market

access in the regional divergence of Italian regions, using the concept of market potential.

This is a measure of the centrality of a region in terms of its access to markets. Given

the lack of trade volumes data for the Italian regions, this measure will be based on the

GDP of each region and the GDP of the adjacent regions, weighted by their distance.

The formulation used here dates back to the seminal work by Harris (1954), adjusted

by the several developments and extensions since then. This paper will estimate the

market potentials of all Italian regions for a series of ten year benchmarks, from 1871 to

1911, following the methodology by Crafts (2005) and Schulze (2007). These estimates

will allow us to look at the market access of different regions both before the process of

industrialization gained ground and during its evolution.

The estimation of market access through market potentials has fruitful applications

beyond the mere quantification of the relative position of the regions. In this paper we

use both domestic and total market potential in order to explain regional GDP per capita

and regional industrial value added per capita. The main result is that regional GDP

per capita in this period was affected by domestic market potential more strongly than

by total market potential, which includes trading partners. In particular, it is the case

4

Figure 3: Industrial value added per capita in the Italian regions, 1871–1911 (constant

1911 prices, Italy=100).

Source: Fenoaltea (2003) and MAIC (1874, 1883, 1902, 1914).

when the model is run in first differences, which corresponds to looking at growth rates.

The second economic indicator that we try to explain is industrial value added per capita,

which in terms of growth rates seems to be much less affected by any type of market

potential. This measure is included in the analysis to help us focus on the part of GDP

generated by the secondary sector, which appears to be more geographically polarized.

This paper is organized as follows. Section 2 discusses the methodology used to

calculate market potentials for the Italian regions; Section 3 illustrates the sources used;

Section 4 shows the estimates and provides a commentary on the results; Section 5 applies

these estimates to explain GDP per capita and industrial value added per capita; Section 6

concludes.

2 Market access in regional analysis: empirical framework

In the literature, market potential has been calculated following several methodologies.

The most basic one goes back to Harris (1954) and uses retail sales weighted by distances

to evaluate the market access of US regions. This type of formulation has been used

5

in more recent works such as Midelfart et al. (2000) on the European Union, Crafts

(2005) on Britain (1870–1910), Schulze (2007) on Austria-Hungary (1870–1910), Martinez-

Galarraga (2012) on Spain (1856–1929) and Klein and Crafts (2012) on the US (1880–

1920). All these works use the total GDP of the regions weighted by distance.2 The

alternative approach to calculating market potentials is to use a gravity model to estimate

the functional form of market potential. A gravity model explains the volumes of trade

among regions using mainly the size of the regions and the distance between each region,

jointly with some dummy variables and controls such as adjacency or the presence of a

border. Market potential in this case is calculated through the parameters estimated by

means of the gravity model. Examples of works using this methodology are Redding and

Venables (2004) at world level, Head and Mayer (2004) on Japanese firms in the European

Union, Hanson (2005) on the United States, and in historical perspective, Wolf (2007) on

Poland (1925–1937) . This exercise is not possible in the case of Italian regions because

data on volumes of trade within the Italian borders are not available for this period. The

next section illustrates the methodology adopted here, which largely follows Crafts (2005)

and Schulze (2007).

2.1 Modelling market potential

In its original formulation, the market potential of region A was defined as the sum of the

GDP of all the adjacent regions, each weighted by their distance from region C, plus the

GDP of region C adjusted by a coefficient that takes account of its size. The calculation

of market potential, following Harris (1954), is shown in Equation (1):

MPc =∑

w

GDPw ×Dγc,w (1)

with Dγc,w the distance between region c and w. The parameter γ is set as -1 and is defined

as in Equation (2):

Dc,c = 0.333×

√

Areacπ

. (2)

The idea behind market potential is quite straightforward. First, for each region C,

we take the main node. The second step is to calculate the distances between the node of

region C and the nodes of each other adjacent region.3 For a given region C, the larger

the GDP of the other regions, the better the access of the regions to markets; the larger

the distance between region C and the other regions and the lower the weight of the GDP

2All the works actually use transport cost adjusted distances; the only exception is Klein and Crafts(2012), which uses straight line distances. The appropriateness and difference between these approachesis discussed later in the paper.

3In this case all the Italian regions plus the main trading partners of Italy.

6

of each of these regions in the market access of the region concerned. Finally, Equation

(1) shows how to deal with the own GDP of the region, which represents the contribution

of the home GDP to the overall market access: Harris (1954) proposes a formula for own

distance that takes into account the size of the region, so that the larger the region, the

lower the weight of its own GDP. The rationale here is that for a given level of GDP, the

larger the region the more spread out, and therefore harder to access, is its own GDP.

Although a gravity model cannot be used because of the lack of internal volumes

of trade data, several refinements are still possible. First of all, distance in our case is

weighted by transport costs as normally done in the literature on market integration and

market potential when volumes of trade are not available. In this case, we decided to take

into account both ground distances, which are assumed to be covered by railway and sea

distances which are assumed to be covered by ship. For each pair of nodes, we calculate

the cheapest combination of railway and shipping. To do so, we apply to all distances both

a variable component (cost per km) and a terminal component (a lump sum cost when

using each given mean of transportation). Whenever a part of the distance is assumed

to be covered with a different mean of transportation, the corresponding terminal cost

is applied. Finally, the last cost to take into account is the existence of trade barriers

between nodes. It is not the case for Italian nodes, but whenever one of the two nodes is

a foreign city, a correction is needed. Following Crafts (2005) and Schulze (2007), tariffs

are converted into distance equivalents. This procedure is based on the coefficients of the

gravity model by Estevadeordal et al. (2002). The elasticities of the model are used to

convert ad valorem tariffs into a distance equivalent measure to be added to the regular

terminal component of the transport cost. The tariff between Italy and each trading

partner is computed as the ratio of the total custom revenues of the trading partner over

its total imports. This gives an average tariff level for each country.

2.2 Testing the effect of market potential on economic development

The goal of this paper is to study the relationship between economic development and

market access in the Italian regions in the period 1871–1911. The empirical framework we

use is taken from the work by Head and Mayer (2011) on market potential and economic

development in the period 1965–2003. This work focuses on the calculation of market

potentials for all countries in the world, relating them to GDP per capita. The main

methodological difference between our work and that by Head and Mayer (2011) is the

calculation of market potentials: here the calculation of market potentials follows Harris

7

(1954) using GDP figures and transport costs; Head and Mayer (2011) use a gravity model

based on trade data.

After obtaining market potential estimates, we implement the model; our goal is to

cast light on the relationship between economic development and market access. We rely

on the following base line specification:

ln(GDPpci)t = βt Market Potentiali

+ αt Region Controlsi

+∑

t

θt Year + ǫt .

(3)

The model described in Equation 3 aims at explaining the GDP per capita of region

i through market potential as the main explanatory variable. The region controls are as

follows: South, which is equal to 1 if the region is in the South (we also show a version of

this model with region fixed effects and with latitude as a control); literacy, which is the

literacy rate in a given region and the share of arable land. This latter is used in 1871

level to explain all years as a method to avoid endogeneity.

Equation 3 can thus be expanded in the following estimating equation:

ln(GDPpci)t = βt Market Potentiali

+ γ1t Latitudei + α1t Literacyi

+ α2t Share Arable Landi

+∑

t

θt Year + ǫt .

(4)

In Section 5 various specifications of this model are shown. We also split the sample

in North and South and we show the same regression with industrial value added per

capita as a dependent variable. All the models are also run in first differences to address

collinearity concerns. Before we show the results, the next section illustrates in detail the

sources used.

8

3 Sources

In this section we describe the sources used for testing the model of Equation 4. All

monetary measures are taken in constant 1911 lire. The regions considered in this work

are the sixteen regions created in 1870 after the annexation of Rome. Figure 4 shows the

boundaries of the Italian regions which did not change over the period.4

Starting the analysis on market potentials in 1871 and ending it in 1911 is both

historically and practically useful. From a pragmatic point of view, the 1871–1911 period

is convenient because borders did not have any variation and because 1871 was the year of

the first census after the main annexations. In fact, Italy was formally unified in 1861 but

its borders changed twice, in 1866 and in 1870, when Veneto and Latium were annexed.

All the main regions of Italy (except Trentino Alto Adige and Venezia Giulia) were part

of Italy in 1871. From an historical point of view, ending the analysis in 1911 allows us

to isolate this period of the early industrialization from the effects of the First World War

and Fascism.

3.1 Market potentials

The variable at the core of this model is market potential. Market potential is calculated

using regional the GDP, the GDP of trading partners, transport cost adjusted distances

and tariffs. The next sections illustrates the sources and how they are used for each of the

components of market potential.

Regional and Foreign GDP Estimates. The first and main ingredient of market

potential is GDP. The estimation of regional disparities for Italy in the period 1871–1911

has been a matter for discussion among scholars for a long time.5 For the GDP estimates

for the Italian regions the latest available series are those from the work of Emanuele

Felice. The estimates of regional GDP used here come from Felice (2009) for 1881 and

1901 and Brunetti et al. (2011), of which Felice is a coauthor, for 1871, 1891 and 1911.

The data provided here on the regional disparities of GDP per capita are the starting point

for deriving GDP estimates in levels for this period. The next step in the procedure is to

apply these per capita disparities to the national GDP per capita. We decided to use for

this step the GDP estimates published by Baffigi (2011) within the broader project of the

Bank of Italy for the 150th anniversary of the Unification. These are the latest estimates

4These regions are quite similar to present regions, with the exception of Venezia Giulia and TrentinoAlto Adige which were not yet Italian and Valle Aosta and Molise which were at the time parts of Piedmontand Abruzzi, respectively.

5See Fenoaltea (2003), Zamagni (1978), Felice (2007), Felice (2009) and Brunetti et al. (2011).

9

Figure 4: Italian regions, 1870–1918.

for the Italian national income and are published both in constant and current prices. In

this paper we use the GDP figures in 1911 constant lire. Starting from the national GDP

estimates by Baffigi (2011), we calculate the national GDP per capita estimates by dividing

them by the present Italian population. This figure is then multiplied by the coefficients

of regional disparities provided by Felice (2009) and Brunetti et al. (2011) to work out all

the level of the regional GDP per capita. Finally, the per capita figures are multiplied by

the regional population figures to obtain the total GDP of each region in levels at constant

1911 prices.6 The regional GDP disparities from Felice (2009) and Brunetti et al. (2011)

are shown in Figure 5.

With regard to the GDP of foreign trading partners, the procedure is the following.

Using data on exports from the Annuario Statistico Italiano, for each benchmark year, we

take all the countries that cover 80% of Italian foreign trade. The trading partners included

are: Austria-Hungary, France, Germany, the United Kingdom, Switzerland, Argentina and

the United States.7 Figure 6 shows the Italian exports to these countries as the share of

total exports for each benchmark year.

6To illustrate, assume that the national GDP per capita in a given year is 100 lire. If the GDP percapita coefficient for a region A is 1.20, the GDP per capita is 120. If the present population of the regionin 1871 is 1,000,000 than the total GDP of region A in 1871 is derived by multiplying these three figures,returning 1,200,000,000 lire.

7The only exception to the 80% criterion is Turkey, for which GDP estimates for this period in currentprices are not easily available from either source.

10

Figure 5: Total GDP of the Italian regions, 1871–1911 (constant 1911 prices, Italy=100).

Source: Felice (2009) for 1881 and 1901 and Brunetti et al. (2011) for 1871, 1891and 1911.

The main source for the GDP of foreign partners is Crafts (2005), who relies primarily

on the work of Prados de la Escosura (2000). Argentina and Switzerland come directly

from Prados de la Escosura (2000) for the years 1881–1911.8 We also consider Austria

and Hungary separately. In order to do so, we split the estimate by Crafts (2005).9

Figure 7 shows the magnitude of the Italian GDP compared to foreign GDP. Looking

at Figure 7, the different magnitude of the US, for 1911 in particular, stands out. Including

such a large GDP compared to that of the Italian regions could be problematic in the sense

that most of the market access could be driven by the US. To underplay the role of the US,

we split it in four macro-regions, Northeast, Midwest, West and South. The implication

of this choice is discussed in more detail in the next section. The regional disparities for

the US are worked out from Klein (2009). These are the same estimates as those on which

the market potentials of Klein and Crafts (2012) are based for 1881–1911. To work out

1871, we use the same disparities as for 1881.

Distances. In order to weight the sum of GDP, a distance measure is required. As in

previous works by Crafts (2005) and Schulze (2007), geographical distance is replaced by

transport costs. This is suitable because straight line distances per se do not take into

account differences of costs for alternative means of transportation nor of the existence

8For 1871, estimates were not available; therefore we used the relative disparity among countries of1881 and applied it to 1871.

9The two GDP are worked out by looking at the relative size in each year from Schulze (2007).

11

Figure 6: Italian export shares, 1871–1911.

Source: Annuario Statistico Italiano 1877, 1881, 1892, 1911.

of railway lines and ports. The first step is to choose a node for each region and foreign

trading partner. Then the distance in terms of railway line or sea route (or the two

together) between each pair of nodes is computed. The administrative regions considered

are shown in Figure 4.

The general rule in selecting the nodes is to take the most populous centre which often

corresponds to the administrative centre. Exceptions are Terni and Pescara for Umbria and

Marches. The reason for this choice is that the actual administrative centres, Perugia and

L’Aquila, were not the economic centres of the regions and were not very well integrated

in its transport network. Using them would have created too high a penalty for Umbria

and Abruzzi in terms of market access. For the US, none of the nodes is the administrative

centre since in the US the largest cities almost never correspond to the capital. The two

means of transportation considered in this period are railways and shipping. For railways,

the length of lines has been worked out from the following sources: Bradshaw’s Continental

Guide of 1914 and the publication on the all lines opened by Ferrovie dello Stato (1927).

Relying on sources that cover the whole period is very important, because this takes into

account the construction of new lines. For shipping, distances were easily computed from

the website www.dataloy.com, which provides the length of maritime routes between all

the main ports worldwide.

Transport Costs. Once a matrix of distances is computed, the next step is to quantify

the rate per tonne per kilometre. The rate taken into account is the average rate between

12

Figure 7: GDP of Italy and its main trading partners, 1871–1911 (constant 1911 millionlire).

Source: Prados de la Escosura (2000) and Crafts (2005).

coal and wheat, which are considered here the two representative goods.10 For Italian

railways, the source is a publication on railway rates by Ferrovie dello Stato (1912). This

publication is quite detailed, providing terminal and variable components of the rate of

transportation for a variety of goods. For the rates of foreign countries, we rely on the

work by Schulze (2007) which uses on the information from the US Bureau for Railway

Economics (1915) and Noyes (1905) to compute terminal and variable components. The

first source provides an overview of 1914 rates for different countries and for both coal

and wheat, separating the cost in terminal and variable component for different city pairs

in each country. From this, the terminal and variable components for various countries

are worked out. The next step is to project these estimates back in time. Noyes (1905)

provides average rates starting from 1870. This information is converted into an index

and used jointly with the 1914 baseline to extrapolate terminal and variable components

for the whole period. For the US, which is not covered by Schulze (2007), we rely on

the information from Noyes (1905) and assume that the US rate is 50% of the general

European rate. For shipping, there are no sources specific to Italy. The estimates for

international ocean shipping from Kaukiainen (2003) are used for all routes. This is a

10We would of course like to take into account the transportation cost of all industrial goods. However,collecting information on transport rate for all goods would be extremely data and time consuming.Moreover, it would not be clear how to use this information in a synthetic measure. Therefore, we followthe existing literature such as Crafts (2005) and Schulze (2007) and adopt the standard solution.

13

Table 1: Regions, trading partners and nodes.

Region/Country Node

Nodes with sea access Liguria Genoa

Venetia Venice

Marches Ancona

Campania Naples

Apulia Bari

Calabria Reggio Calabria

Sicily Palermo

Sardinia Cagliari

United Kingdom London

Turkey Istanbul

Argentina Buenos Aires

Unites States New York

Northeast New York

South New Orleans

West San Francisco

Nodes without access to the sea Piedmont Turin

Lombardy Milan

Emilia Bologna

Tuscany Florence

Umbria Terni

Latium Rome

Abruzzi Pescara

Basilicata Potenza

Austria-Hungary Vienna

France Paris

Germany Berlin

Switzerland Zurich

widely used source for this type of research, including works on Italy.11 The transport

costs used are set out in Tables 2, 3 and 4.

Table 2: Shipping rate per tonne per km, in constant 1911 lire, 1871-1911.

Terminal Component Cost per km

1871 19.67 0.00307

1881 14.71 0.00180

1891 11.14 0.00116

1901 9.06 0.00111

1911 7.29 0.0011

Source: Kaukiainen (2003)

11Federico (2007) uses the same source to study the market integration of Italy in the 19th century.

14

Table 3: Italian railway rates per tonne per km, in constant 1911 lire, 1871-1911.

Terminal Component Cost per km

1871 1.93 0.05688

1881 1.83 0.05386

1891 1.51 0.04456

1901 1.56 0.04588

1911 1.26 0.03731

Source: Ferrovie dello Stato (1912), Schulze (2007), Noyes (1905)

Table 4: Railway and Shipping rates per tonne per km, in constant 1911 lire 1871-1911.

Austria Railway Rates France Railway Rates Germany Railway Rates

Terminal Variable Terminal Variable Terminal Variable

1871 5.44 0.08342 7.67 0.01937 6.56 0.03344

1881 4.35 0.06633 8.69 0.02195 6.69 0.03379

1891 3.11 0.04860 7.61 0.01924 6.21 0.03149

1901 2.98 0.04618 7.00 0.01753 5.96 0.03057

1911 2.60 0.04013 5.73 0.01450 4.90 0.02505

Europe Railway Rates UK Railway Rates US Railway Rates

Terminal Variable Terminal Variable Terminal Variable

1871 7.55 0.034289 2.35 0.052741 7.99 0.03614

1881 7.58 0.034409 2.34 0.05054 4.88 0.022046

1891 6.32 0.028692 2.04 0.045535 3.63 0.016437

1901 6.19 0.028006 2.29 0.0491 3.10 0.014003

1911 5.10 0.023171 1.87 0.041229 2.55 0.011585

Source: Ferrovie dello Stato (1912), Schulze (2007), Noyes (1905) andUS Bureau for Railway Economics (1915)

Tariffs. The last cost to be taken into account is the one originated by trade barriers

between nodes. It is not the case for Italian nodes, but whenever one of the two nodes is

a foreign city, a correction is needed. Following Crafts (2005) and Schulze (2007), tariffs

are converted into distance equivalents. The source used is Mitchell (2003) except for the

United Kingdom, where Mitchell (1988) is used. Capie (1994) gives data for 1870 Germany

and Ferreres (2005) for Argentina. Whenever a year is missing, either the closest available

year is used or the gap is filled by interpolation. Table 5 shows the level of tariffs used.

3.2 Region controls

In Section 5, GDP per capita and industrial value added per capita are explained using

market potentials, geographical controls (such as dummies for macro areas, region fixed

effects and latitude) and two other controls; literacy rates and share of arable land. The

share of arable land is shown in Figure 8 while literacy rates in Figure 2 are derived from

(A’Hearn et al., 2011).

15

Table 5: Ad valorem tariffs, 1871–1911.

Austria-Hungary France Germany UK Switzerland Argentina US

1871 2% 3% 8% 6% 2% 22% 29%

1881 0% 6% 6% 5% 2% 27% 24%

1891 3% 8% 9% 5% 3% 20% 20%

1901 6% 9% 9% 9% 4% 29% 20%

1911 6% 10% 8% 12% 4% 2% 15%

Source: Mitchell (2003) for Austria-Hungary, France and Germany, Mitchell (1988)for the United Kingdom, Capie (1994) for 1870 Germany and Ferreres (2005) forArgentina.

Figure 8: Share of arable land by province, 1870.

Source: MAIC (1976).

A last remark is on the year 1891. The 1891 population census has not been carried

out owing to budget restrictions. For this reason, explanatory variables such as literacy

rates are not available. In order to avoid relying heavily on interpolations, we decided to

leave 1891 out of the regression analysis. The next section shows the results of the market

potential calculations.

4 Market potential of Italian regions, 1871–1911: empirical

results

This section shows and compares different versions of market potential for the Italian

regions that will be used in the next section as explanatory variable for GDP per capita and

industrial value added per capita. The first one proposed is a domestic market potential,

exclusively taking into account the Italian regions. The second one is a repetition of the

domestic market potential using straight distances and shows how different results can

16

Figure 9: Domestic market potential in Italian regions, 1871–1911 (constant 1911 prices,Italy=100).

Source: our own calculations.

emerge without controlling for transport costs. The third is total market potential, which

comprises all the main trading partners of Italy along with the Italian regions.

Let us start with the first version. Figure 9 and Table 6 provide the estimates of

domestic market potential for the Italian regions between 1871 and 1911.

The domestic market potential shows two main results. The first is that at the

beginning of the period the picture is quite in line with the classic North-South divide,

with the North showing a higher level of market access and the South lagging behind. The

exceptions in the South are Campania and Sicily. The reason why these two regions have

levels comparable to the Northern regions is that their total GDP at the beginning of the

period were comparable to those of the regions in the North and that these two regions

have sea-ports as their economic centre. This makes them able to exploit shipping, which

is cheaper than railways in this period. The second result is that there is a tendency

over the period for the market access of the North to worsen with respect to the South.

This tendency can be explained if we look at the transport costs in Tables 2, 3 and 4.

Shipping costs drop relatively faster in this period than railway costs, giving an advantage

to regions that have their node on the coast. This is the case of the main Southern regions.

A similar result, connected to the access to shipping, is discussed in Schulze (2007) on the

Habsburg Empire. In its case, the only two regions with access to the sea (Littoral and

Dalmatia) have a persistent advantage over the others in terms of market potential in spite

17

Table 6: Domestic market potential in the Italian regions, 1871–1911.

1911 million lire Italy=100

1871 1881 1891 1901 1911 1871 1881 1891 1901 1911

Piedmont 766 932 1264 1528 2552 120 118 114 113 109

Liguria 768 946 1385 1702 2973 121 119 123 126 127

Lombardy 908 1033 1496 1779 3094 142 130 133 131 132

Veneto 777 893 1274 1570 2716 122 113 114 116 116

Emilia 783 889 1270 1414 2550 123 112 114 104 109

Tuscany 733 863 1197 1340 2365 115 109 106 99 101

Marches 612 774 1115 1374 2343 96 98 99 101 100

Umbria 541 593 811 916 1588 85 75 72 68 68

Latium 560 728 1013 1182 2063 88 92 89 87 88

Abruzzi 509 608 826 951 1634 80 77 74 70 70

Campania 747 915 1298 1574 2712 117 115 116 116 116

Apulia 547 765 1105 1401 2352 86 96 98 103 101

Basilicata 336 536 746 874 1489 53 68 67 64 64

Calabria 512 701 998 1307 2234 80 88 89 96 96

Sicily 641 869 1228 1527 2593 101 110 108 113 111

Sardinia 457 642 952 1247 2139 72 81 85 92 91

Italy 637 793 1124 1355 2337 100 100 100 100 100

Source: our own calculations.

of their lower GDP. This relationship between sea access and market potential is clear when

trading partners are taken out of the sample. This is because Littoral and Dalmatia are

the only regions in the sample directly connected by sea to the foreign trading partners.

Taking trading partners out, the two regions no longer have such advantage. In Italy this

advantage is present even without the inclusion of trading partners because shipping is an

available option in internal trade as well.

The second set of estimates uses straight line distances between nodes. Figure 10

and Table 7 show the domestic market potentials when we take straight line distances

instead of transport costs. Here we see that the process of worsening market access

worsening in the North and improving in the South is not taking place. This version

of market access also shows a much wider gap between North and South, with most

Southern regions doing sensibly worse than is implied by the calculation adjusted for

transport costs. This version of market potential is the most similar to the one proposed

by A’Hearn and Venables (2011). The authors calculate market potentials using the

same GDP estimates and weight them by straight line distances. Their results show a

large gap between North and South: the authors claim that domestic market access was

the driving force for the location of industries in the period 1890–1950. Looking at the

18

Figure 10: Domestic market potential in Italian regions, 1871–1911 (constant 1911 prices,Italy=100, straight line distances).

Source: our own calculations.

difference between the domestic market potentials calculated with transport costs adjusted

distances and with straight line distances, the claim that market potentials were moving

in the same direction as GDP or industrial value added appears to be supported much

less by the former calculation.12

The third version proposed here, which we call total market potential, shows the market

potentials calculated including the main trading partners of Italy.13 Figure 11 and Table

8 show the results.

This version of market potential shows a quite different picture from the domestic one.

The South here appears not only to perform increasingly better in the period but also

as starting from a higher level than the North, which experiences the opposite evolution.

These estimates, which pool the GDP of Italian regions and that of very large trading

partners, are very much influenced by the GDP of the latter, in particular the US, which

has a very high GDP compared to Italy (see 12). Given that most trading partners are

reached at least partially by sea, the sea effect is even higher in this formulation of market

potential. Taking the US as one unit would force us to pick only one node and weight

the entire GDP of the US according to this node, which in our case is New York. This

12If we compared total market potential calculated with transport costs and with straight line distancesthe difference would look even more marked.

13The choice of trading partners to include is discussed in Section 3 above.

19

Table 7: Domestic market potential in the Italian regions, 1871–1911 (straight linedistances).

1911 million lire Italy=100

1871 1881 1891 1901 1911 1871 1881 1891 1901 1911

Piedmont 61 74 81 95 143 116 123 120 128 121

Liguria 70 78 95 106 177 134 130 138 142 150

Lombardy 78 86 105 121 194 149 144 154 163 164

Veneto 61 62 71 82 131 116 104 105 110 111

Emilia 73 81 97 103 172 140 135 142 138 146

Tuscany 63 71 81 85 139 120 119 118 114 117

Marches 47 53 61 65 102 90 89 90 88 86

Umbria 48 53 59 63 98 92 89 86 84 83

Latium 51 63 71 77 124 99 105 101 104 105

Abruzzi 45 52 56 59 93 87 86 82 80 78

Campania 67 73 82 85 134 129 122 121 114 113

Apulia 38 49 55 59 89 74 82 80 80 75

Basilicata 35 42 46 49 75 68 70 69 66 63

Calabria 31 38 40 43 67 60 64 59 58 57

Sicily 44 55 61 64 98 84 93 87 87 83

Sardinia 22 27 31 34 53 43 45 45 45 45

Italy 52 60 68 74 118 100 100 100 100 100

Source: our own calculations.

practically corresponds to assuming that the whole GDP of the US is as easily accessible

as if it was all located on the East Coast. This of course is far from the fact. This is why

we decided to consider the US as four separate macro regions (Northeast, Midwest, South

and West). Doing so allows us to reduce the role of the US in the calculation without

ignoring the presence of the US in the foreign markets.14

The next section move our focus to the relationship between these estimates and

regional economic development.

14The same problem could be posed by Austria and Hungary: we decided to keep them separate in allcalculations as in this period, it must be remembered that the two regions of the Habsburg Empire becamea dual monarchy in 1867 and therefore were quite economically independent.

20

Figure 11: Total market potential in Italian regions, 1871–1911 (constant 1911 prices,Italy=100).

Source: our own calculations.

Table 8: Total market potential in the Italian regions, 1871–1911.

1911 million lire Italy=100

1871 1881 1891 1901 1911 1871 1881 1891 1901 1911

Piedmont 3,976 4,739 7,020 10,609 15,071 108 93 88 85 86

Liguria 4,322 5,912 9,361 14,902 20,886 118 116 117 119 119

Lombardy 4,010 5,032 7,697 11,533 16,665 109 99 96 92 95

Venetia 3,900 5,506 8,775 13,763 19,136 106 108 110 110 109

Emilia 3,402 4,312 6,589 9,744 14,216 93 84 82 78 81

Tuscany 3,628 4,874 7,523 11,470 16,421 99 95 94 92 93

Marches 3,673 5,363 8,582 13,573 18,790 100 105 107 108 107

Umbria 3,028 4,136 6,408 9,744 13,895 82 81 80 78 79

Latium 3,452 4,877 7,601 11,812 16,775 94 96 95 94 95

Abruzzi 3,048 4,235 6,577 10,086 14,293 83 83 82 81 81

Campania 4,025 5,791 9,188 14,600 20,372 109 113 115 117 116

Apulia 3,673 5,445 8,709 13,869 19,193 100 107 109 111 109

Basilicata 2,934 4,230 6,588 10,177 14,418 80 83 82 81 82

Calabria 3,767 5,548 8,846 14,251 19,767 102 109 111 114 112

Sicily 4,159 6,036 9,515 15,404 21,391 113 118 119 123 122

Sardinia 3,837 5,646 9,027 14,640 20,340 104 111 113 117 116

Italy 3,677 5,105 8,000 12,511 17,602 100 100 100 100 100

Source: our own calculations.

21

Figure 12: Contribution of trading partners to total market potential, 1871–1911.

Source: our own calculations.

22

5 Market access and economic development: empirical

results

Section 4 presented market potential estimates for the Italian regions for five benchmark

years, between 1871 and 1911. In this section we explore the relationship between

market potential and regional GDP per capita, which represents a fundamental measure

of economic development. We then show the same model using industrial value added

per capita as the dependent variable, which represents the part of the GDP of a region

that comes from the secondary sector. To do so, we use both basic pooled OLS and first

difference OLS. The main explanatory variable in all specifications is market potential,

along with other geographic and economic controls. In the next section we show the results

on GDP per capita.

5.1 Market potential and GDP per capita

In this section we show the empirical results based on the model of section 2.2. Table 9

shows the OLS estimates of Equation 4. All years are pooled and standard errors are

heteroskedasticity-robust.15 Columns 1 and 2 show the basic pooled OLS regression with

GDP per capita explained by domestic and total market potential. Here there are no

controls. The relationship appears positive and significant at the 1% level for domestic

market potential as well as for total market potential. The coefficients are expressed in

logs and therefore they can be interpreted as elasticities. Domestic market potential has

an elasticity of 0.351 and total market potential has an elasticity of 0.235. The R2 is

about 10% higher for the specification with domestic market potential, which is near 0.5,

suggesting that the former has more explanatory power. The comparison with previous

literature appears challenging, because the existing works are at country level rather than

sub-national level. Redding and Venables (2004, p. 65) provide a coefficient of 0.146 for

domestic market potential and 0.395 for total market potential. Head and Mayer (2011,

p. 289) find a coefficient of 0.80 for the specification without specific country controls.

As admitted by the authors, the difference in the magnitude of the coefficients is more

than double that of Redding and Venables (2004) and “this difference in the coefficients

stems mainly from the different construction of the market potential variable”. Therefore,

differences in the type of sample, in the historical periods and in the techniques used to

calculate market potentials allow for very different magnitudes in the coefficients without

affecting the validity of the different works. In columns 3 and 4 of Table 9 we add year

15The pooling of the four available years (1871, 1881, 1901 and 1911) is necessary in order to increasethe number of observations. This procedure is the one followed by Head and Mayer (2011).

23

dummies to take into account the fact that the data are pooled across years. In this

version, the results change: domestic market potential increases its elasticity to 0.544

significant at 1% while total market potential is not significant in this case. The next step

is to introduce further controls to capture differences across regions. The first candidate is

physical geography. In columns 5 and 6 we introduce a dummy for belonging to different

macro areas.16 In the regression, we decided to use the North-West as our baseline and

include the dummies corresponding to the other two macro areas. The result is that

the domestic market potential decreases its coefficient size and level of significance while

the total market potential moves in the opposite direction. Both columns have market

potentials significant at the 5% level while the dummy South is highly significant with

a negative sign. The North-East-Centre is also negative but with a considerably smaller

coefficient. The negative signs on both macro areas are not surprising, since we chose the

richest macro area as the baseline. The fact that being in the South has the strongest

effect is also expected because of the generally poorer conditions of Southern regions in

this period. The use of these macro areas should be considered in light of the types of

difference we want to capture. If the aim is to describe geographic differences across

Italy, avoiding other elements that could be endogenous to GDP, the imposition of these

macro borders is not necessarily the best option in spite of the strong results. This is

because macro areas are often designed ex-post to collect regions with similar economic

conditions.17 In order to avoid this endogeneity problem, columns 7 and 8 use latitude

as the geographic control. Latitude is a control that is completely exogenous, being the

distance from the Equator, and has a positive effect on GDP per capita. The result is that

both coefficients are around 0.4, with domestic market potential significant at the 1% and

total market potential at 5%. The R2 decreases slightly compared to that in columns 6

and 7, but are very much higher than the specification without geographic controls. From

now on, whenever we decide to include a control for the geographic position of the regions

we will use latitude.

The previous table aimed at exploring the relationship between market potential and

GDP per capita using controls that are possibly not endogenous to GDP per capita.

In Table 10 we try to introduce further controls and discuss whether or not these are

appropriate in our case. The first control that we can think of including, to capture

16The classification of regions in the three different macro areas is taken from Felice (2007) and was laterused in his other works. The Northwest includes Piedmont, Lombardy and Liguria; the Northeast-Centreincludes Venetia, Emilia, Marches, Tuscany, Umbria and Latium; the South comprises Campania, Apulia,Basilicata, Calabria, Sicily and Sardinia.

17See for example the inclusion of the North-East in the same macro area of the Centre and the decisionto keep the regions of the Industrial Triangle separate from the rest.

24

Table 9: GDP per capita and market potential, 1871–1911 (pooled OLS regression withgeographic controls).

GDPpc (1) (2) (3) (4) (5) (6) (7) (8)

Log Domestic MP 0.351*** 0.544*** 0.217** 0.431***

(0.0452) (0.0853) (0.0939) (0.0929)

Log Total MP 0.235*** 0.191 0.306** 0.411**

(0.0450) (0.194) (0.115) (0.172)

Northeast-Centre -0.142** -0.163***

(0.0566) (0.0597)

South -0.315*** -0.389***

(0.0535) (0.0406)

Latitude 0.0207** 0.0452***

(0.00801) (0.00961)

Constant -1.189 0.768 -5.049*** 1.773 1.760 -0.529 -3.634** -4.983

(0.941) (1.022) (1.732) (4.275) (1.920) (2.545) (1.792) (4.070)

Year Fixed Effects no no yes yes yes yes yes yes

Region Fixed Effects no no no no no no no no

Observations 64 64 64 64 64 64 64 64

R2 0.491 0.309 0.537 0.376 0.667 0.671 0.563 0.519

Notes: Heteroskedastic robust standard errors in parentheses. * , ** and ***correspond to a coefficient significantly different from zero with a 10%, 5% and 1%confidence level respectively. The dependent variable is the log of GDP per capita in1911 prices. Market potentials are in 1911 prices.

other fundamental variables that can affect GDP per capita is some measure of human

capital. In modern studies, such as Head and Mayer (2011), average years of schooling

is the best variable to use. For the case of pre-First World War Italy we decided to use

a more basic output measure that is often used to capture human capital in the Italian

case: literacy rates. When literacy rates are introduced in the regression, they are highly

significant with a coefficient of around 0.6 in both specifications. Market potentials loose

their significance and latitude is incorrectly signed. These results clearly suffer from a bias.

The reason why the plain inclusion of this variable is problematic is two-fold: literacy is

endogenous to GDP per capita and it is also collinear to other variables that present a

North-South gradient.18 This is, for instance, the case of domestic market potential, which

has a geographic distribution that appears very similar to that of literacy rates. In this

case, it is very hard to disentangle the effect of literacy from that of market potential;

moreover, it is very hard to establish the direction of the causality between literacy and

GDP per capita. Columns 3 and 4 show the inclusion of the share of arable land in 1871

as a control.19 Including arable land at the beginning of the period is a way to avoid

18We also tried to included similar controls such as the share of labour force in agriculture; the problemsencountered in the estimation were very similar and we decided to deal with literacy rates only in theestimation to avoid making the issue even more problematic by including further controls.

19The share of arable land is also used as a control by Redding and Venables (2004). The authorshere use a number of other controls, such as fraction of land in tropical areas, prevalence of malaria and

25

Table 10: GDP per capita and market potential, 1871–1911 (pooled OLS regression withfurther controls).

Log GDPpc (1) (2) (3) (4) (5) (6) (7) (8)

Log Domestic MP -0.0531 0.431*** 0.430*** 0.430**

(0.125) (0.0843) (0.127) (0.181)

Log Total MP -0.155 0.335** -0.206 -0.206

(0.158) (0.162) (0.212) (0.286)

Log Literacy 0.623*** 0.640***

(0.118) (0.0998)

Latitude -0.0522*** -0.0595*** 0.0241*** 0.0468***

(0.0148) (0.0178) (0.00825) (0.00961)

Log Share of Arable Land 1871 -0.192*** -0.153**

(0.0513) (0.0627)

Liguria 0.165** 0.245*** 0.165*** 0.245***

(0.0665) (0.0816) (0.0127) (0.0695)

Lombardy -0.0766 0.00284 -0.0766** 0.00284

(0.0773) (0.0588) (0.0280) (0.0181)

Venetia -0.262*** -0.223*** -0.262*** -0.223***

(0.0690) (0.0766) (0.00276) (0.0451)

Emilia -0.153** -0.184*** -0.153*** -0.184***

(0.0680) (0.0488) (0.00469) (0.0282)

Tuscany -0.0989* -0.129*** -0.0989*** -0.129***

(0.0517) (0.0459) (0.0149) (0.00718)

Marches -0.269*** -0.308*** -0.269*** -0.308***

(0.0508) (0.0454) (0.0273) (0.0366)

Umbria 0.00337 -0.188*** 0.00337 -0.188***

(0.0659) (0.0489) (0.0682) (0.0411)

Latium 0.388*** 0.283*** 0.388*** 0.283***

(0.0605) (0.0433) (0.0467) (0.00728)

Abruzzi -0.269*** -0.483*** -0.269*** -0.483***

(0.0763) (0.0638) (0.0796) (0.0345)

Campania -0.175*** -0.127* -0.175*** -0.127**

(0.0610) (0.0704) (0.00215) (0.0597)

Apulia -0.128* -0.174** -0.128*** -0.174***

(0.0705) (0.0659) (0.0318) (0.0407)

Basilicata -0.201** -0.494*** -0.201* -0.494***

(0.0922) (0.0483) (0.112) (0.0360)

Calabria -0.366*** -0.437*** -0.366*** -0.437***

(0.0649) (0.0599) (0.0443) (0.0480)

Sicily -0.204*** -0.177** -0.204*** -0.177**

(0.0631) (0.0720) (0.0106) (0.0723)

Sardinia -0.153** -0.251*** -0.153** -0.251***

(0.0596) (0.0752) (0.0575) (0.0545)

Constant 7.257*** 9.838** -3.100* -2.845 -2.631 10.66** -2.631 10.66

(2.631) (3.856) (1.564) (3.856) (2.570) (4.661) (3.689) (6.287)

Year Fixed Effects yes yes yes yes yes yes yes yes

Region Fixed Effects no no no no yes yes yes yes

Region Clustering no no no no no no yes yes

Observations 64 64 64 64 64 64 64 64

R2 0.760 0.763 0.599 0.541 0.955 0.948 0.955 0.948

Notes: Heteroskedastic robust standard errors in parentheses. * , ** and ***correspond to a coefficient significantly different from zero with a 10%, 5% and 1%confidence level respectively. The dependent variable is the log of GDP per capita in1911 prices. Market potentials are in 1911 prices. Literacy is expressed as rate.

26

endogeneity as the share of arable land in subsequent years may have been affected by

changes in the economic or technological conditions of the regions. The share of arable

land is significant and negatively signed in both specifications. The negative sign can be

explained by interpreting the share of arable land as the presence of agricultural activity

in a region. Since agriculture has lower value added levels than industries, leading to lower

levels of GDP per capita, it has a larger weight in the regional economy. Here latitude is

correctly signed and significant again, suggesting that endogeneity and multicollinearity

are less of an issue in this specification. Columns 4–8 take a different approach to the

specification. In the last four columns we introduce region fixed effects. This strategy is

also used by Head and Mayer (2011). However, in their case the inclusion of these dummies

leads to much less of an increase in the R2 than our case shows. Here, region fixed effects

lead to an increase from about 0.5 to values well over 0.90. This increase is explained by

the number of regional dummies that are highly significant. The baseline is Piedmont, a

fairly rich Northern region. This is why most other significant regional dummies have a

negative sign (with the exceptions of Liguria and Latium which do show quite high levels

of GDP per capita even compared to Piedmont in this period). Southern regions all have

negative signs, which is expected given the lower levels of GDP per capita in this part of

the country. The high explanatory power of these fixed effects suggests that the inclusion

of any further control other than market potential should be dropped in order to avoid

the risk of over-identifying the model. For this reason, given the high share of variation

explained, we assume that the region fixed effects capture a sufficient part of the regional

differences (although they are not able to disentangle them) and we do not include further

controls. Columns 6 and 7 show the inclusion of fixed effects without regional clustering,

while Columns 8 and 9 show the results with clustering. Market potential here shows

similar results to the previous specification: domestic market potential is positive and

significant with an elasticity of 0.430; total market potential is not significant.

Summing up the results of the first two tables, we notice that domestic market potential

is significant and correctly signed in all specifications except when literacy rates are

introduced while total market potential is not significant in a number of specifications,

most notably when fixed effects are used, appearing less robust as the explanatory variable

for GDP.

We now move to the issues of endogeneity and multicollinearity that affect the model

when literacy rates are introduced. On the first point, Head and Mayer (2011) attempt

risk of expropriation that appear sensible when doing a cross-country analysis but would lead to difficultquantification or have little meaning in the case of Italian regions in this period.

27

two different approaches. First, they substitute total market potential with foreign market

potential, which is market potential calculated with own GDP removed.20 This solution

has been attempted for the case of the Italian regions but the results do not hold.

According to Head and Mayer (2011, p. 291),“[foreign market potential] has nice features

[in dealing with endogeneity], but is clearly not ideal as a replacement or instrument

for RMP”. The solution that Head and Mayer (2011) adopt in their work is to use an

instrument for market potential. In particular, they take geographic centrality proxied

by two instruments: the sum of the inverse of straight line distances and the inverse of

transport costs. In their case the former is ruled out and the solution adopted is to use the

latter. Unfortunately, this instrument does not appear to work for the Italian regions in

this period. Looking at the levels of our variables, in the absence of reliable instruments,

we cannot establish the direction of causality between GDP per capita, domestic market

potential and literacy. This is evident when looking at the geographical distribution of the

levels in Figures 1, 2 and 10, where the North-South gradient appears clear. One solution

here is to move our focus from levels to changes in the time of the variables of interest.

By transforming all the logs of the variables in first differences we can effectively interpret

the model in terms of growth rates. Running the model in first differences allows us to

address the issue of multicollinearity, since the common long-term trend in the variables

is not present in the growth rates.21

Table 11 shows the basic formulation without region or geographic controls in Columns

1 and 2. The result is a positive and significant coefficient for the domestic market potential

and very low and insignificant coefficients on the total market potential.22 Columns 3–8

replicate the main specifications proposed so far: latitude as control with literacy rates first

and then with the share of arable land and, in the last two columns, region fixed effects

as only controls. Unlike the results with levels, first difference regressions show little

effect from the controls, while domestic market potentials are always correctly signed and

significant with a coefficient around 0.8 in all the specifications. Total market potential

does not appear significant in any specification.

20This leaves us effectively with a market access measured only accounting for the GDP of foreigncountries in the sample, ignoring the internal market.

21In particular, if we look at Figures 1, 2 and 10, we notice that literacy rates and domestic marketpotential in the South tend to converge with those of the North while GDP per capita tends to diverge.

22The coefficients here can be interpreted as percentage point increases in growth rate of GDP percapita when an explanatory variable increases by 1%. Therefore, in column 1, if the growth rate ofdomestic market access increases by 1%, the growth rate of GDP per capita increases by 0.649%.

28

Table 11: GDP per capita, market potential and literacy rates, 1871–1911 (pooled OLSregression in first differences).

Log GDPpc (1) (2) (3) (4) (5) (6) (7) (8)

DLog Domestic MP 0.649*** 0.849*** 0.816*** 0.883**

(0.206) (0.235) (0.249) (0.345)

DLog Total MP 0.0557 0.0731 0.111 0.370

(0.332) (0.399) (0.373) (0.489)

DLog Literacy 0.0645 -0.245

(0.272) (0.304)

Latitude 0.0124 -0.00195 0.0112* 0.00296

(0.00812) (0.0102) (0.00612) (0.00761)

Share of Arable Land 1871 -0.000347 -0.00146

(0.00109) (0.00127)

Constant -0.0764 0.258** -0.724 0.392 -0.629* 0.167 -0.133 0.216

(0.113) (0.114) (0.449) (0.575) (0.335) (0.395) (0.190) (0.129)

Year Fixed Effects yes yes yes yes yes yes yes yes

Region Fixed Effects no no no no no no yes yes

Observations 48 48 48 48 48 48 48 48

R2 0.619 0.520 0.651 0.531 0.651 0.532 0.696 0.592

Notes: Heteroskedastic robust standard errors in parentheses. * , ** and ***correspond to a coefficient significantly different from zero with a 10%, 5% and 1%confidence level respectively. The dependent variable is the log of GDP per capita in1911 prices. Market potentials are in 1911 prices. Literacy is expressed as rate. Allvariables are in first differences.

5.2 Market potential in the North and South

Tables 12 and 13 show our model splitting the sample in the North (which merges

the North-West and North-East-Centre) and the South.23 As a result, the number of

observations goes down to 21 for the South and 27 for the North. Let us start with the

South. In Table 12 we notice that the level of GDP per capita in Southern regions is

much more clearly affected by market potential when the regression is in levels (Columns

1–4) than in first differences (Columns 5–8). This is the case both without geographic

controls and when we include latitude and share of arable land (Columns 3–4). Moreover,

the coefficients are slightly higher for total market potential than for the domestic one.

In first differences, the results do not hold, suggesting that after Unification market

potentials do nothing to explain the growth rates of GDP per capita.

The picture looks quite different in Table 13, where the sample is restricted to

the northern regions. In this case, total market potential is never significant while

domestic market potential is significant in both levels and first differences, with fairly

23Venetia, Lombardy, Piedmont, Liguria, Emilia, Tuscany, Latium, Umbria and Marches are consideredNorth; the rest of the regions are considered South as in footnote 16.

29

Table 12: GDP per capita and market potential, southern regions, 1871–1911 (pooled OLSregression and first differences).

Log GDPpc (1) (2) (3) (4) (5) (6) (7) (8)

Log Domestic MP 0.514*** 0.642***

(0.0647) (0.0996)

Log Total MP 0.624*** 1.030***

(0.104) (0.176)

Latitude 0.0361** 0.0589*** 0.00476 0.00591

(0.0146) (0.0164) (0.0111) (0.0106)

Log Share of Arable Land 1871 -0.132** 0.0298

(0.0539) (0.0508)

DLog Domestic MP 0.617** 0.489

(0.260) (0.302)

DLog Total MP 1.316 0.976

(0.914) (1.033)

DLog Literacy 0.539 0.427 0.523 0.446

(0.437) (0.419) (0.455) (0.435)

Share of Arable Land 1871 -0.00205 -0.00291

(0.00205) (0.00221)

Constant -4.486*** -7.906*** -8.046*** -19.31*** -0.227 -0.303 -0.269 -0.324

(1.295) (2.266) (2.203) (4.336) (0.212) (0.324) (0.539) (0.608)

Year Fixed Effects yes yes yes yes yes yes yes yes

Observations 28 28 28 28 21 21 21 21

R2 0.824 0.750 0.866 0.870 0.700 0.652 0.719 0.697

Notes: Heteroskedastic robust standard errors in parentheses. * , ** and ***correspond to a coefficient significantly different from zero with a 10%, 5% and 1%confidence level respectively. The dependent variable is the log of GDP per capitain 1911 prices. Market potentials are in 1911 prices. Literacy is expressed as rate.Columns 5–8 show the variables in first differences.

high coefficients. Market potential is therefore a stronger predictor of GDP per capita

within northern regions compared to the South.

Considering the North and South of Italy separately brings some interesting insights.

The effect of market access on the GDP per capita of these two different parts of the

country is not the same: in the South, both domestic and total market potential have a

strong role in determining the levels of GDP per capita but they do not seem to impact

on the growth rates. For the North, only the domestic market potential is significant in

both in levels and in first differences. This difference suggests that the results at national

level are driven by the Northern part of the country, where the access to markets outside

Italy was often more expensive because of the presence of rich but landlocked regions.

The next section moves from GDP to industrial production, repeating the exercise

using industrial value added per capita as the dependent variable.

5.3 Market potential and industrial value added per capita

The growth of the industrial output is considered a strong driver of economic growth for

Italy in this period. One further question we can address is how far the factors that explain

30

Table 13: GDP per capita and market potential, northern regions, 1871–1911 (pooled OLSregression and first differences).

Log GDPpc (1) (2) (3) (4) (5) (6) (7) (8)

Log Domestic MP 0.120 1.458***

(0.210) (0.212)

Log Total MP 0.194 0.0767

(0.275) (0.259)

Latitude -0.221*** -0.0517 0.00255 0.0000621

(0.0366) (0.0325) (0.0181) (0.0283)

Log Share of Arable Land 1871 -0.318*** -0.400***

(0.0889) (0.107)

DLog Domestic MP 1.552*** 1.586***

(0.338) (0.343)

DLog Literacy -0.241 -0.109 -0.356 -0.176

(0.291) (0.354) (0.451) (0.584)

DLog Total MP -0.146 -0.128

(0.449) (0.515)

Share of Arable Land 1871 0.00152 0.000694

(0.00206) (0.00238)

Constant 3.632 1.811 -12.75*** 8.108 -0.530*** 0.355** -0.698 0.332

(4.298) (6.063) (3.062) (5.572) (0.179) (0.167) (0.813) (1.352)

Observations 36 36 36 36 27 27 27 27

R2 0.450 0.453 0.804 0.633 0.762 0.569 0.770 0.570

Notes: Heteroskedastic robust standard errors in parentheses. * , ** and ***correspond to a coefficient significantly different from zero with a 10%, 5% and 1%confidence level respectively. The dependent variable is the log of GDP per capita in1911 prices. Market potentials are in 1911 prices. Literacy is expressed as rate.

GDP per capita can also explain industrial value added per capita. Let us start with

Table 14, in which the main OLS regressions in levels are reported. Columns 1 and 2 show

the baseline model with year dummies as the only controls. Here domestic market potential

is positive and significant at 1% while total market potential is significant at the 5% level.

The elasticities are considerably higher than the ones for GDP per capita. Columns 3 and

4 introducing the dummy for latitude show the same result; Columns 5 and 6 report the

specification with the share of arable land and latitude together. It appears that in all

specifications, both domestic and total market potentials are positive and significant with

high coefficients (between 0.6 and 1) and a 1% level of significance. However, the results

do not hold when fixed effects are introduced. This last two Columns provide a counter

intuitive result for total market potential, which is negative and significant. To investigate

further, we turn our attention to Table 15, where the same exercise is repeated in first

difference. Here again the results on market potentials do not hold. As in the case of the

GDP per capita of Southern regions, market potential seems to possibly explain levels of

GDP per capita but not the changes from period to period. The next section sums up the

results.

31

Table 14: Industrial value added per capita and market potential, 1871–1911 (pooled OLSregression).

Log Ind VA pc (1) (2) (3) (4) (5) (6) (7) (8)

Log Domestic MP 1.136*** 0.997*** 0.997*** -0.369

(0.142) (0.153) (0.151) (0.240)

Log Total MP 0.651** 1.059*** 1.022*** -0.835***

(0.272) (0.207) (0.214) (0.214)

Latitude 0.0255** 0.0840*** 0.0289** 0.0847***

(0.0110) (0.0132) (0.0116) (0.0135)

Log Share of Arable Land 1871 -0.193*** -0.0741

(0.0656) (0.0885)

Constant -18.96*** -10.30* -17.22*** -22.84*** -16.68*** -21.80*** 11.84** 22.62***

(2.897) (5.989) (2.950) (4.779) (2.842) (4.963) (4.895) (4.707)

Year Fixed Effects yes yes yes yes yes yes yes yes

Region Fixed Effects no no no no no no yes yes

Observations 64 64 64 64 64 64 64 64

R2 0.801 0.538 0.817 0.742 0.832 0.744 0.953 0.957

Notes: Heteroskedastic robust standard errors in parentheses. * , ** and ***correspond to a coefficient significantly different from zero with a 10%, 5% and 1%confidence level respectively. The dependent variable is the log of the industrial valueadded per capita in 1911 prices. Market potentials are in 1911 prices. Literacy isexpressed as rate.

Table 15: Industrial value added per capita, market potential and literacy rates, 1871–1911(pooled OLS regression in first differences.

Log Ind VA pc (1) (2) (3) (4) (5) (6) (7) (8)

DLog Domestic MP -0.0138 0.378 0.281 0.406*

(0.195) (0.251) (0.249) (0.216)

DLog Total MP -0.291 0.268 0.275 0.565*

(0.253) (0.304) (0.281) (0.323)

Latitude 0.0244*** 0.0223*** 0.0248*** 0.0241***

(0.00615) (0.00658) (0.00605) (0.00614)

Share of Arable Land 1871 -0.00268* -0.00307**

(0.00144) (0.00136)

Constant 0.389*** 0.481*** -0.980*** -0.888** -0.722* -0.621* 0.141** 0.311***

(0.107) (0.0878) (0.299) (0.343) (0.362) (0.329) (0.0519) (0.0849)

Year Fixed Effects no no yes yes yes yes yes yes

Region Fixed Effects no no no no no no yes yes

Observations 48 48 48 48 48 48 48 48

R2 0.483 0.492 0.622 0.603 0.655 0.648 0.796 0.792

Notes: Heteroskedastic robust standard errors in parentheses. * , ** and **correspond to a coefficient significantly different from zero with a 10%, 5% and 1%confidence level respectively. The dependent variable is the log of industrial valueadded per capita in 1911 prices. Market potentials are in 1911 prices. Literacy isexpressed as rate. All variables are in first differences.

32

6 Conclusions

This paper presented the estimates of market potentials of the Italian regions for 10-year

benchmarks in the period 1871–1911 and used market potential to explain GDP per capita

and industrial value added per capita. Market potentials are based on constant 1911 price

estimates of regional GDP for Italy from Felice (2009) and Brunetti et al. (2011) and the

GDP of the main trading partners of Italy in the period from 1871 to 1911 are from Prados

de la Escosura (2000) and Crafts (2005). In the calculation of market potentials, the GDP

are weighted by distance-adjusted transport costs to take into account the actual distance

among regions in terms of the costs of transporting one unit of a representative good.

In the case of foreign partners a distance-equivalent tariff is calculated and added to the

transport cost. In this paper we proposed different specifications ranging from domestic

market potential, which comprises Italian regions only to total market potential that

includes all Italian regions and all the main trading partners. The bottom line of these

calculations is the following: domestic market potentials at the beginning of the period

show a more traditional picture of Italy, with the North presenting higher values than the

South, in particular at the beginning of the period. What contradicts the North-South

traditional image is that North and South converge. This is explained by the relative

cost of shipping, which goes down, unlike that of railways, over the period. This is an

advantage for the main Southern regions such as Campania and Sicily, which have good

access to the sea along with a fairly high total GDP, while Piedmont and Lombardy for

instance do not. The role of transport costs is easily appreciated when we compare the

same calculation with straight line distances: the picture is reversed, showing the strong

and consistent position of the Northern regions. These comparisons show how useful it

is, at least for the case of Italy and other countries with more than one viable mean of

transportation, to correctly account for transport costs and existing lines. Introducing

foreign markets in the total market potential, the results change for the first year, but the

trend in time goes in the same direction. The North has no advantage in terms of market

access at the beginning of the period and, as for the domestic one, it worsens its position

compared to the South.

Once the estimates of market potentials have been produced, we used them as

explanatory variables for the main measure of economic performance: GDP per capita.

We first proposed a baseline specification in levels and we then added further controls. In

the specifications in levels, we found that domestic market potential has a stronger and

more significant positive effect on GDP per capita than total market potential. These

33

findings are in line with our expectations from the descriptive analysis of the data.

However, the inclusion of economic controls such as literacy rates cause concerns over

the multicollinearity and endogeneity that could be affecting the results. To deal with

these issues, we ran the model in first differences. Taking the first differences of the logs

corresponds to considering the variables in growth rates. This procedure leaves out the

common trends that the variables might present in the long run and restricts the focus to

the change from one benchmark year to another. The model in first differences confirms

that the domestic market potential is positive and significant when trying to explain the

growth rate of GDP per capita of the Italian regions.

The next step was to divide the sample into North and South, running the model

separately. It is quite useful to look at different parts of the country separately whenever

the number of observations allows us to do so. In this case it is especially useful to separate

levels and growth rates as the main result here is that total market potential determined

the level of GDP per capita in the South but it did not in the North. Moreover, growth

rates in market potential of both types do not affect the growth rates of GDP per capita

in the South but they do in the North when we consider domestic market potential. This

insight tells us that in the South, the total market potential was positively correlated to

the levels of GDP per capita but its growth rate did not drive the growth rate of GDP

per capita. In the rest of the country, both the level and the growth rates of total market

potential are not significant in explaining GDP per capita. This suggests that the access

to international markets did matter more for the South compared to the rest of Italy.

This result is in line with the view that the Southern economy was, until the invasion of

agricultural products from the US in the late 19th, quite open to international trade. A well

known study by Morilla et al. (1999, pp. 333–337) takes into account the case of southern

Italy as exporter of high value added agricultural products. The authors give the example

of citrus production in Sicily, which supplied 95% of lemons and 16% of oranges consumed

in the US in 1890. However, the trend over the period before the First World War was

negative for the southern Italian exporters because of the competition from California. We

believe that this decline in exports is informative on the first difference regressions results

on total market potential. Finally, we ran the model in levels and first differences, using

industrial value added per capita as the dependent variable. Here the main result was

that both domestic and total market potentials were positive and significant in explaining

the levels of industrial value added per capita, but the results were not robust to the use

of fixed effect and first differences. This suggests that market potential, in its domestic

34

formulation, is a more clear predictor of GDP per capita than industrial value added per

capita.

The findings of this paper contribute to the existing literature in two ways. First of

all, the paper suggests that the quantification of market access can be quite sensitive to

the inclusion of trading partners as well as to the proper quantification of transport costs

rather than simple geographical distances. Furthermore, the different role of domestic

and international market potentials challenges the idea that the South of Italy has been

penalized by its geographical position in terms of access to the international markets as

suggested for instance by Daniele and Malanima (2007).

35

Primary Sources

[1] Bureau of Railway Economics, Comparison of railway freight rates in the United

States, the principal countries of Europe, South Australia, and South Africa.

Washington, 1915.

[2] Ferrovie delo Stato (FS), Tariffe e condizioni pei trasporti sulle Ferrovie dello Stato.

Roma, 1912.

[3] Ferrovie delo Stato (FS), Sviluppo delle ferrovie italiane dal 1839 al 31 dicembre

1926. Roma, 1927.

[4] Istituto Nazionale di Statistica (ISTAT), Annuario statistico italiano, various years.

[5] Ministero di agricoltura, industria e commercio (MAIC), Relazione sulla condizione

dell’agricoltura nel quinquiennio 1870–1874. Roma, 1876

[6] Ministero di agricoltura, industria e commercio (MAIC), Censimento generale al 31

dicembre 1871. Roma, 1874.

[7] Ministero di agricoltura, industria e commercio (MAIC), Censimento della

popolazione del Regno d’Italia al 31 dicembre 1881. Roma, 1883.

[8] Ministero di agricoltura, industria e commercio (MAIC), Censimento della

popolazione del Regno d’Italia al 10 febbraio 1901. Roma, 1902.

[9] Ministero di agricoltura, industria e commercio (MAIC), Censimento della

popolazione del Regno d’Italia al 10 giugno 1911. Roma, 1914.

References

A’Hearn, B. (2000): “Could Southern Italians Cooperate? Banche Popolari in the

Mezzogiorno,” Journal of Economic History, 60, 67–93.

A’Hearn, B., C. Auria, and G. Vecchi (2011): In ricchezza e in poverta, Il Mulino,

Bologna, chap. Istruzione.

A’Hearn, B. and A. Venables (2011): “Internal Geography and External Trade:

regional disparities in Italy, 1861-2011,” Economic History Working Papers 12, Bank of

Italy.

Baffigi, A. (2011): “Italian National Accounts, 1861-2011,” Economic History Working

Papers 18, Bank of Italy.

Brunetti, A., E. Felice, and G. Vecchi (2011): In ricchezza e in poverta, Il Mulino,

Bologna, chap. Reddito.

Cafagna, L. (1989): Dualismo e sviluppo nella storia d’Italia, Marsilio, Padova.

Capie, F. (1994): Tariffs and Growth: Some Illustrations from the World Economy,

1850-1940, Manchester University Press, Manchester.

Crafts, N. (2005): “Market potential in British regions, 1871-1931,” Regional Studies,

39, 1159–1166.

Daniele, V. and P. Malanima (2007): “Il prodotto delle regioni italiane e il divario

nord-sud in Italia, 1861-2004,” Rivista di Politica Economica, 47, 267–315.

Estevadeordal, A., B. Frantz, and A. M. Taylor (2002): “The Rise and Fall of

World Trade, 1870-1939,” Working Papers 9318, NBER.

Felice, E. (2007): Divari regionali e intervento pubblico: per una rilettura dello sviluppo

in Italia, Il Mulino, Bologna.

——— (2009): “Estimating regional GDP in Italy (1871-2001): sources, methodology and

results,” Working Papers in Economic History 0907, Universidad Carlos III.

——— (2013): “Regional income inequality in Italy in the long run (1871-2001). Patterns

and determinants,” UHE Working Papers 08, Universitat Autonoma de Barcelona.

37

Felice, E. and M. Vasta (2012): “Passive Modernization? The new Human

Development Index and its Components in Italy’s Regions (1871-2007),” UHE Working

Papers 10, Universitat Autonoma de Barcelona.

Fenoaltea, S. (2003): “Peeking Backward: Regional Aspects of Industrial Growth in

Post-Unification Italy,” Journal of Economic History, 63, 1059–1102.

——— (2006): L’economia italiana dall’Unita alla Grande Guerra, Collezione storica,

Laterza, Bari.

Ferreres, O. (2005): Dos siglos de economıa argentina, 1810-2004: : historia argentina

en cifras, Fundaciøn Norte y Sur, Buenos Aires.

Galassi, F. (2000): “Measuring Social Capital: Culture as an Explanation of Italy’s

Economic Dualism,” Department of economic working papers, University of Warwick.

Hanson, G. (2005): “Market potential, increasing returns and geographic concentration,”

Journal of International Economics, 67, 1–24.

Harris, C. D. (1954): “The Market as a Factor in the Localization of Industry in the

United States,” Annals of the Association of American Geographers, 44, 315–348.

Head, K. and T. Mayer (2004): “Market Potential and the Location of Japanese

Investment in the European Union,” Review of Economics and Statistics, 86, 959–972.

——— (2011): “Gravity, market potential and economic development,” Journal of

Economic Geography, 11, 281–294.

Kaukiainen, M. (2003): “How the price of distance declined: ocean freights for grain and

coal from the 1870s to 2000,” Department of history, University of Helsinki, MIMEO.

Klein, A. (2009): “Personal Income of U.S. States: Estimates for the Period 1880–1910,”

Economics Research Paper Series (TWERPS) 916, University of Warwick.

Klein, A. and N. Crafts (2012): “Making sense of the manufacturing belt:

determinants of U.S. industrial location, 1880–1920,” Journal of Economic Geography,

12, 775–807.

Martinez-Galarraga, J. (2012): “The determinants of industrial location in Spain,

1856–1929,” Explorations in Economic History, 49, 255–275.

38

Midelfart, K., H. Overman, S. Redding, and A. Venables (2000): “The

Location of European Industry,” European Economy Economic Papers 142, European

Commission.

Missiaia, A. (2014): “Industrial location, market access and economic development:

regional patterns in post-unification Italy,” PhD Thesis, London School of Economics

and Political Science.

Mitchell, B. (1988): British Historical Statistics, Cambridge University Press,

Cambridge.

——— (2003): International historical statistics: Europe, 1750-2000, Palgrave Macmillan,

New York.

Morilla, C., A. Olmstead, and P. Rhode (1999): “Horn of Plenty: The

Globalization of Mediterranean Horticulture and the Economic Development of

Southern Europe, 1880-1930,” Journal of Economic History, 59, 316–352.

Noyes, W. (1905): American railroad rates, Little, Brown and Company, New York.

Prados de la Escosura, L. (2000): “International Comparisons of Real Product, 1820-

1990: an Alternative Data Set,” Explorations in Economic History, 37, 1–41.

Redding, S. and A. J. Venables (2004): “Economic geography and international

inequality,” Journal of International Economics, 62, 53–82.

Schulze, M. (2007): “Regional income dispersion and market potential in the late

nineteenth century Hapsburg Empire,” Working Paper 106-07, London School of

Economics.

Sonnino, S. (1877): La Sicilia nel 1876, Barbera, Firenze, chap. I contadini in Sicilia.

Villari, P. (1979): Le Lettere Meridionali ed Altri Scritti sulla Questione Sociale in

Italia, Guida, Napoli.

Wolf, N. (2007): “Endowments vs. market potential: What explains the relocation of

industry after the Polish reunification in 1918?” Explorations in Economic History, 44,

22 – 42.

Zamagni, V. (1978): Industrializzazione e squilibri regionali in Italia: bilancio dell’eta

giolittiana, Il Mulino, Bologna.

39

——— (1990): Dalla periferia al centro: la seconda rinascita economica dell’Italia (1861-

1990), Il Mulino, Bologna.

40